The addition of lockdep code to write_seqcount_begin/end has lead to
a bunch of false positive claims of ABBA deadlocks with the so_lock
spinlock. Audits show that this simply cannot happen because the
read side code does not spin while holding so_lock.